Edit | History | Changes Home page | Site map | Search | Recent changes | Help

Columbia-hw-2004-06-11-comments
Columbia Comments July 9, 2004 on UMd and CMU IL2s, and Other Issues

Did Columbia analysis before styding CMU & UMd in detail.

Assumptions in Columbia analysis

* Only fields in IL2: concept and Sem relation -- set TrEd optionsto not show anythingbut these slots!!

Issues revealed during exercise:

* Need for dealing with discourse-style relations

* Need for scope: can (not (do X)) vs not( can (do X))

* "problems such as .... are experienced" vs "not satisfied with ..." are not meaning-preserving in the sense of IL2 (only using real0-world knowledge)

* What to do with attributes?

Compatibiliy with CMU:

* Largely compatible with Simon's analysis, though differences in choices of how to represent certain discourse relations.

Compatibility with UMd:

Here are some fairly superficial comments on Rebecca's analysis.

Format:

* Rebecca has a dag (or her structures may in general be a directed graph). Notation is somewhat hard to read; suggestion: draw graphs by hand and scan and post (I drew the graph out in 2 dimensions to understand it).

Note: maybe a graph is a reinterpretation of what Rebacca did, but I think not, it is just another representation.

* Rebecca has typed nodes -- events, states, objects.

* Objects do not have a concept associated with them (why?).

Issue of Graph vs. Tree

* Coreference: non-issue. Whether there are 3 ndoes marked as coindexed or one node pointed to by three arrows makes no difference. Note: I did not think to mark event co-reference, which of course is crucial.

* Some questions about headedness (also with CMU/Columbia, but more here). For example, is the state/event of the PERMITTED state 1 the LOCATION state or the SERVICE-PROVISION state?

Content:

* Rebecca makes exlicit the first switch, which is alluded to by "again" -- so she has 3 switches, CMU and Columbia only 2. She also makes explicit the SERVICE-PROVISION states, and reprepsent knoeldge about them. CMU and Columbia do not have these at all.

Some Specific, Minor Issues:

* Aren't in fact the switches permitted or not permitted, not the service provision itself?

* The switches have no agents, but that is crucial here, also wrt the permission -- a switch may be permitted if initiated by the govenment, or the companies.


Version 2, Wed 09 Jun 2004 20:05:29 [OCR] - created Wed 09 Jun 2004 20:04:07 [OCR]
Edit | History | Changes Home page | Site map | Search | Recent changes | Help